Hello, A new project on the LCRC cluster has been requested. Please forward the information on to the LCRC Allocation sub-committee. Applicant's name: Maria Chan Applicant's institution: ANL Applicant's division: NST Project Name: GATEM Project title: Genetic Algorithm Optimization of Interfacial Structures from Electron Microscopy Associated funding: LDRD Other Systems: None Science: Electronic, optical, mechanical, electrochemical, and piezoelectric properties are significantly influenced by the presence and characteristics of solid-solid interfaces such as grain boundaries. Atomic structure of materials can be characterized by transmission electron microscopy (TEM) and scanning TEM (STEM). However, it is very challenging to characterize interfaces and surfaces due to imperfections and defects at these regions. The goal of this project is to establish a quantitative method for comparing simulated S/TEM images with experimental S/TEM images, so that comparative decisions about image similarity rely on a quantitative algorithm, rather than expert judgment. Successful automation of a similarity decision, along with energetic information from empirical potential or DFT, is a crucial step in developing a multi-objective genetic algorithm optimization loop for determining 3D structures at a disordered interface. Due to the iterative n ature of the overall optimization process, emphasis must be placed on making each step computationally efficient. Current image simulation techniques based on an FFT multislice algorithm are used to qualitatively confirm the interpretation of conventional TEM images formed from coherently scattered electrons, and are relatively efficient, as they assume a periodic input structure and operate within the constraints of the weak phase object approximation. We will extend the current image simulation capabilities to non-periodic specimens in a computationally efficient manner and include signals from high angle annular dark field (HAADF) region to achieve high lateral resolution, Z-contrast imaging. Project description: A crucial component of most recognition-based tasks in the field of computer vision (CV) is the definition of “image similarity” relative to a target image. This work adapts a popular CV framework, scale invariant feature transform (SIFT), for the purposed of measuring the similarity between real and simulated atomic resolution microscopy images. The similarity metric (SIFT descriptor) will be developed offline, but requires several simulated/experimental image matches to use as a test set for training a machine learning model and/or algorithm refinement. Once the similarity metric is proven adequate, the effort will then shift to parallelizing the current multi-slice code for use in a multi-objective genetic algorithm optimization method, aimed at determining the full three-dimensional structure of materials at solid-solid interfaces. The multi-objective genetic algorithm will be used to optimize the atomic structure to obtain optimum image simi larity from the constructed SIFT descriptor, and the energy of interfaces, simultaneously. The energy evaluations of interfaces can be accurately computed with DFT methods, whereas empirical potentials provide also a reasonably accurate thermodynamics with significant reduction in computational cost. We will start with computing the interface energies using the available empirical potentials for CdTe, namely Stillinger-Weber (Z. Q. Wang, D. Stroud, and A. J. Markworth, Physical Review B 40, 3129 (1989).) and Bond Order potential (D. K. Ward, X. W. Zhou, B. M. Wong, F. P. Doty, and J. A. Zimmerman, Physical Review B 85, 115206 (2012).) from molecular dynamics (MD) simulations in LAMMPS (http://lammps.sandia.gov/). All image simulations will be carried out using the Kirkland TEMSIM code (C++), available as open source from https://sourceforge.net/projects/computem/. The breakdown of the required calculations is as follows: (i) Descriptor development: A single simulated image requires about 500 core hours of computation. At this stage, we require 10 fully-converged simulated images for a total of 10*500 = 5,000 core hours to refine the current SIFT adaptation. Convergence is achieved when the total integrated intensity of the detector region is nearly unity, which is a function of pixel size. For imaging areas in 40‎Å x 40‎Å window, a 1024x1024 sampling size is typical. We will run quick 1D line scans on a few of the images before proceeding with constructing the set for the descriptor development. This convergence checking procedure will only need to be carried out once and will apply to the remaining calculations. It is estimated that this will require 5,000 core hours. (ii) TEMSIM parallelization: Efforts are in order to adapt the Kirkland code (E. J. Kirkland. Advanced Computing in Electron Microscopy. Plenum, New York, (1998)) to a parallel computing environment by propagating several probes at the same time on different processors, and/or by parsing the input structure and distributing each unique structure division to its own core. With two parallelization schemes to test, this overall effort is estimated to require 25,000 core hours. (iii) Multi-objective GA optimization: For the image simulation component of the initial GA testing, we anticipate iterating through 100 generations with 6 images per generation. This calculation requires 100*6*500 = 300,000 core hours for the image simulation component. The time estimated for each MD simulation is 250 core hours. In total, 100 generations * 6 images/structures * 250 core hours/structure = 150,000 core hours. Total computation time requested: 5,000 + 5,000 + 25,000 + 300,000 + 150,000 = 485,000 core hours. Industry partnership: Project URL: Requested allocation: 485000 Q1: 125000 Q2: 120000 Q3: 120000 Q4: 120000 Justification: Storage requirements: The requester has used undetermined amount hours of their initial startup project. In addition to approving an initial amount, please specify a Category and Subcategory for this project. For a list of the current selection of approved categories, please see: https://wiki.lcrc.anl.gov/wiki/Processes/Categories Once the Allocation committee has approved the project, please go to the Project Management page to create it: https://accounts.lcrc.anl.gov/projects.php Thank You, The LCRC Accounts System